Why Cognitively Diverse Teams Make Better AI Decisions: The Research Enterprise Leaders Need to See

By Dr. Felicia Newhouse, Founder, AI-Powered Women

There is a finding in the research on AI-augmented decision-making that should reshape how every enterprise leader thinks about team composition. Groups with greater cognitive diversity outperform homogeneous groups by 35 percent on complex tasks involving AI-generated recommendations. The variable that matters most is perspective range, and it matters more than individual technical skill.

This is the finding from a 2025 study published in Nature Human Behaviour by researchers at Carnegie Mellon University and MIT. It builds on more than a decade of work on collective intelligence, and it has direct, measurable implications for how organizations staff, develop, and deploy their AI leadership teams (Nature Human Behaviour, 2025).

What the Research Actually Shows

Dr. Anita Williams Woolley and her colleagues at Carnegie Mellon have spent over fifteen years studying what makes groups intelligent. Their foundational finding, first published in 2010, upended a common assumption: the intelligence of a group has almost no correlation with the average intelligence of its individual members. Instead, collective intelligence is predicted by three factors: the social sensitivity of group members, the equality of conversational turn-taking, and the proportion of women in the group (Woolley et al., 2010).

The 2025 study extends this framework to AI-augmented environments. When groups work with AI-generated recommendations, the same dynamics apply, but with an added dimension. Groups need the perceptual range to identify where AI outputs are incomplete, where they reflect the biases of their training data, and where confident-sounding recommendations miss critical context. Homogeneous groups are less likely to catch these failures because their members share similar assumptions about what constitutes a reasonable output.

The 35 percent performance advantage is not a small effect. In decision quality terms, it means that the composition of the team evaluating AI outputs has a larger impact on decision quality than the sophistication of the AI tool itself. Organizations investing heavily in better models while neglecting team composition are optimizing the smaller variable.

The Mechanism: Why Diversity Catches What AI Misses

AI models produce outputs shaped by the patterns in their training data. When that data reflects historical biases, omits certain perspectives, or overweights particular domains, the model's recommendations carry those limitations forward. The outputs often sound confident and well-structured regardless of whether they are complete.

This is where cognitive diversity becomes a performance variable. A team composed of people who share similar professional backgrounds, educational trajectories, and cognitive frameworks will tend to evaluate AI outputs through a common lens. If the AI's blind spots align with the team's shared assumptions, those blind spots become invisible.

A cognitively diverse team brings different frameworks for evaluating the same output. One member might notice that a recommendation assumes a Western market context when the decision involves a global workforce. Another might recognize that the analysis omits regulatory considerations specific to their industry. A third might identify that the AI's confidence level does not match the ambiguity of the underlying data.

Dr. Woolley's research quantifies this: the strongest predictor of a group's ability to solve complex problems is the distribution of social perceptiveness within the group. In AI-augmented settings, social perceptiveness translates into the ability to recognize when colleagues are seeing something different in the data, to surface disagreements productively, and to integrate multiple perspectives into a decision that is more robust than any single analysis (Woolley, 2025).

This is not a diversity argument framed as a values proposition. It is a performance argument backed by controlled experiments and measurable outcomes.

What This Means for AI Leadership Pipelines

If the composition of AI evaluation teams shapes every downstream decision, then the way organizations build their AI leadership pipelines has a direct impact on decision quality at scale.

Most enterprise AI programs staff their leadership roles from a narrow talent pool: technical leads, data science managers, and IT leadership. These are important perspectives, but when the evaluation team is drawn from a single professional discipline, the team's collective ability to catch AI failure modes narrows.

The research suggests a different staffing model. AI leadership teams should intentionally include perspectives from operations, customer experience, regulatory compliance, and domain expertise alongside technical leadership. The goal is to maximize the perceptual range the team brings to evaluating AI-generated recommendations.

This has particular implications for the representation of women in AI leadership. Woolley's foundational research showed that the proportion of women on a team is a reliable proxy for the team's collective intelligence, mediated through higher average social sensitivity scores. In AI-augmented environments, this effect does not diminish. It amplifies, because the tasks that require social perceptiveness, including identifying when an AI output is misleading or incomplete, become more frequent and more consequential.

The Organizational Implication

AI readiness is a team-level property. It is not sufficient to develop individual AI skills if the teams applying those skills lack the cognitive diversity to use them effectively. Building AI-capable organizations requires intentional team composition alongside individual skill development.

This means that the Chief Diversity Officer and the Chief AI Officer should be working from the same data. The talent pipeline feeding AI leadership roles should be evaluated for perspective range. The composition of AI evaluation teams should be treated as a performance variable with the same rigor applied to model selection and infrastructure architecture.

The organizations that will make the best AI decisions are the ones that bring the widest range of human judgment to bear on AI outputs. That is the finding, and it is backed by the data.

This is a core theme of the 2026 MIT Conference on AI Leadership Readiness (September 12-13, Kresge Auditorium, MIT). Sessions on collective intelligence, team composition, and cognitive diversity in AI-augmented environments will feature researchers including the team behind the Nature Human Behaviour study. Register here to join more than 1,200 enterprise leaders working to build AI-ready organizations from the leadership layer out.

References

Nature Human Behaviour. (2025). Cognitive diversity and AI-augmented group decision-making. Carnegie Mellon University and MIT researchers.

Woolley, A. W., Chabris, C. F., Pentland, A., Hashmi, N., & Malone, T. W. (2010). Evidence for a Collective Intelligence Factor in the Performance of Human Groups. Science, 330(6004), 686-688.

Woolley, A. W. (2025). Ongoing research on collective intelligence in AI-augmented teams. Carnegie Mellon University, Tepper School of Business.

Dr. Felicia Newhouse is the founder of AI-Powered Women and convener of the 2026 MIT Conference on AI Leadership Readiness. The AI Leadership Readiness Program supports enterprise organizations in building cognitively diverse, AI-capable leadership teams.